Do brand mentions get you into AI answers? What the data shows

What is a brand mention?

A brand mention is your product’s name in a third-party article, review or list, with or without a link. The pages that matter most for software companies are the ones buyers already read before they shortlist: comparison roundups, review lists and “alternatives to X” pages.

AI answer engines such as ChatGPT, Perplexity and Google AI Overviews draw on that same text when a buyer asks which tool to use. That is the reason mentions have moved from a PR metric to something a link building programme has to plan for.

What did the Ahrefs study find?

In Ahrefs’ study of 75,000 brands, published in May 2025 (Linehan, 2025), branded web mentions correlated with AI Overview visibility at 0.664. Backlinks correlated at 0.218. On a scale where 1 would be a perfect match, mentions were a much stronger companion to visibility than links were.

We treat this as the strongest signal in the largest public test we know of. We do not treat it as a promise, and we do not sell it as one.

Does a correlation mean mentions cause citations?

No, and this is the point where most articles on the topic go wrong. Brands that are mentioned a lot are usually brands that are already well known, and well-known brands are also the ones AI engines tend to name. Mentions and visibility can rise together because both follow from being established.

There’s no public data yet proving that a mention causes a citation. We track it, we don’t promise it. That sentence is what we say to every client before a brand mentions campaign starts.

Which pages do AI engines draw from?

Not the same pages everywhere. In our own tests and in the research we work from (Husky Hamster, 2026), the source preferences differ by platform:

• Gemini leans towards government, education and medical references.
• Perplexity cites e-commerce and comparison sites heavily.
• ChatGPT leans towards Reddit, YouTube and forums.

Citation overlap between ChatGPT, Gemini and Perplexity is small, only 12 to 17%. A placement that helps on one platform tells you little about the others, so we map targets per platform instead of assuming one list works for all of them.

How do you work backwards from the answers?

Start from what buyers ask, not from what you can buy. Our brand mentions process runs in six steps:

1. Prompt set. Agree the questions your buyers actually ask an AI engine.
2. Baseline. Record which brands each answer names and which pages it cites.
3. Target. The cited pages become the shortlist: comparison lists, review roundups, “alternatives to X” pages.
4. Qualify and place. Every site passes the same rules as any other placement. The mention says only what is true of your product, in a place a reader would expect to find it.
5. Re-check. Run the same prompts again and report where you are named and where you are not.
6. Monitor. Live, indexed and unchanged, checked for 12 months. A removed placement is replaced free.

Placements are $295 each, flat, content included. The full service is described on the brand mentions page.

How do you measure whether it worked?

Not with a single screenshot. In the research we use (Husky Hamster, 2026), half of the number-one positions in AI answers were unstable after 30 minutes, and 72% had changed after four weeks. A one-off position report is therefore a snapshot, not a measurement.

What holds up is presence over time: at least four runs a month on 100 or more prompts, reported as the percentage of answers that name you and how that set rotates. Track the intent separately too. Definition questions were the most stable in that research, at about 59%, while recommendation questions were the least, at about 31%. A brand that appears for “what is X” is on firmer ground than one that appears for “best X for my team”.

What can go wrong?

Three things, in the order we see them:

• Volume in one intent. A hundred identical mentions aimed at one question do not build visibility. Vary the intents and contexts instead, using a gap matrix of competitor, intent and funnel stage.
• Fake reviews. Detection of inconsistent reviews keeps improving, and one penalised account can erase months of work. We use authentic reviews only.
• Bought links. Where a publisher charges for a placement, we pay it. Placements carry no marks: we don’t ask publishers to label them as sponsored or add link attributes. Search engines can devalue bought links, and you carry that risk with us. We tell you which placements involved a fee before you approve, and you approve every site.

Where should a SaaS team start?

With five questions your buyers ask an AI engine, written down exactly as they would type them. Run them, note which pages the answers cite, and check whether your product is named on those pages. That list is a better plan than any number of placements bought without it.